DETERMINING THE CAUSAL LINKS BETWEEN REPRODUCTIVE TIMING AND METABOLIC RISK: A BIDIRECTIONAL MENDELIAN RANDOMIZATION STUDY
BMIN5030 Final Project
Author
Pauline Gachanja
1Overview
This project uses Mendelian Randomization (MR) to investigate the bidirectional relationship between reproductive timing, particularly age at menarche (AAM) and age of natural menopause (ANM) with Type 2 Diabetes and Lipids (HDL, LDL, Triglycerides and Total cholesterol). The goal is to determine whether we can use reproductive timing as a diagnostic tool to predict Type 2 Diabetes risk and lipid levels or vice-versa.
2Introduction
The prevalence of metabolic traits differs largely between men and women. Men tend to have higher incidences throughout their life, leading to the underrpresentation of women in genetic research, based on the assumption that women are protected against metabolic disorders. However, in reality, the mortality associated with metabolic and related cardiometabolic diseases is higher in women than in men. The risk of developing these diseases increases after menopause, a transition that has been linked to the loss of endogenous estrogen. Estrogen deficiency contributes to metabolic dysfunction through mechanisms including endothelial dysfunction, insulin resistance, and dyslipidemia. Both age at menarche (AAM) and age at natural menopause (ANM) reflect key hormonal transitions across a woman’s life course and are therefore important biological markers of long-term metabolic health.
In this study, I am going to use genome-wide association studies (GWAS) summary statistics from large, publicly available studies to investigate the causal direction between each reproductive timing (age at natural menopause and age at menarche) and metabolic risk (Type 2 Diabetes and lipids [HDL, LDL, Triglycerides and Total cholesterol]) using Mendelian Randomization (MR). MR integrates genetics and causal inference, using genetic variants as instrumental variables to determine causation between an exposure and the outcome variable by minimizing confounding causation and reverse causation. A bi-directional MR will be implemented to evaluate whether reproductive timing influences metabolic disease risk and conversely, whether metabolic factors influence the variation of reproductive timing. In addition, a multivariable MR analysis will be conducted by introducing another variable BM1, and determining whether the exposure influences the outcome independent of BMI and vice-versa. The findings from this project aim to evaluate whether reproductive timing can serve as a potential early-life screening marker for identifying women at increased risk of adverse metabolic outcomes and to clarify the central role of BMI in these relationships.
Data sources
To avoid confounding factors and bias, the datasets used for MR should be independent (no-overlapping individuals) of each other and from the same ancestry.
Table 1: Summary of the datasets used in the MR analysis
Relevance: The genetic variants (SNPs) used as instruments must be strongly associated with the exposure. This is assessed by GWAS significance threshold (p < 5×10⁻⁸) and F-statistics.
This is essential for the instrument to meaningfully predict the exposure
No confounders: The genetic instruments should not be associated with the confounders of the exposure-outcome relationship.
No Horizontal Pleiotropy: The genetic instruments must influence the outcome only through the exposure and not through any other biological pathway.
Study design used for this project
Forward Mendelian Randomization (AAM or ANM -> T2D or Lipid Levels)
Reverse Mendelian Randomization (T2D or Lipid Levels -> AAM or ANM )
Multivariable Mendelian Randomization for both forward and reverse MR, where we include BMI as a confounding exposure variable.
3.2 Main statistical Methods
F-statistics: This is done to determine whether the genetic instruments are strongly associated with the exposure. If the F-statistics is greater than 10, then the genetic instruments are assumed to be strong, and if they less than 10, then they are considered to be weak and this would introduce bias.
Clumping: This removes SNPs in linkage disequilibrium (LD) with the causal variant, to ensure each SNP provides independent information about the exposure.
Harmonization: This is done to ensure that the exposure and the outcomes SNP effects correspond to the same allele. This step ensures that the direction of effect is consistent across datasets and prevents sign errors that could lead to incorrect causal estimates. Palindromic SNPs with ambiguous strand orientation are handled or removed to maintain accuracy.
MR estimates
Inverse Variance Weighted (IVW): This calculates the weighted sum of the wald ratios (beta of the exposure/ beta of the outcome) across all SNPs. It is the primary method and most accurate as it gives unbiased results especially when the assumptions of MR are held. I have based my interpretation based on this method.
MR-Egger: It allows for and is sensitive to directional horizontal pleiotropy by estimating an intercept term. It provides a valid causal estimate even when pleiotropy is present, though at the cost of reduced statistical power.
Weighted Median: It produces a consistent causal estimate if at least 50% of the total instrument weight comes from valid SNPs. It is robust to violations of assumptions affecting a portion of the instruments.
Weighted Mode: This method identifies the most common (modal) causal estimate among SNPs, assuming that the largest subset of instruments contributing to the same causal effect are valid.
Simple Mode: The simple mode estimator identifies the mode of the causal effect distribution without applying weights. It is less precise than the weighted mode but useful as an additional sensitivity method.
Sensitivity Analyses
Heterogeneity (Cochran’s Q): It checks whether the SNPs have a consistent causal effect. If the p-value <0.05 then heterogeneity is present, indicating possible directional horizontal pleiotropy or presence of invalid instruments.
Steiger directionality test: Checks whether the assumed direction in MR is correct. Determines whether the SNPs explains more variance in the exposure or the outcome, if it explains more in the exposure then the direction is correct. If the p-value <0.05 then the direction of MR is correct.
MR-Egger intercept pleiotropy test: Checks for directional horizontal pleiotropy. if significant then there is pleiotropy which disrupts the MR assumptions and could lead to causal effect bias. If the p-value <0.05 then there is pleiotropy
Leave-one-out analysis: It takes out one SNP at a time and re-calculates the causal effect. This helps to determine potential outliers.
Single SNP analysis: It calculates the causal estimate per SNP without combining them, to help identify outliers.
TwoSampleMR version 0.6.25
[>] New authentication requirements: https://mrcieu.github.io/ieugwasr/articles/guide.html#authentication.
[>] Major upgrades to our servers completed to improve service and stability.
[>] We need your help to shape our emerging roadmap!
Please take 2 minutes to give us feedback -
https://forms.office.com/e/eSr7EFAfCG
Warning:
You are running an old version of the TwoSampleMR package.
This version: 0.6.25
Latest version: 0.6.26
Please consider updating using remotes::install_github('MRCIEU/TwoSampleMR')
library(ieugwasr)
OpenGWAS updates:
Date: 2024-05-17
[>] OpenGWAS is growing!
[>] Please take 2 minutes to give us feedback -
[>] It will help directly shape our emerging roadmap
[>] https://forms.office.com/e/eSr7EFAfCG
Attaching package: 'ieugwasr'
The following object is masked from 'package:TwoSampleMR':
ld_matrix
library(readr)library(dplyr)library(remotes)
Warning: package 'remotes' was built under R version 4.3.3
Attaching package: 'remotes'
The following object is masked from 'package:TwoSampleMR':
add_metadata
library(devtools)
Loading required package: usethis
Attaching package: 'devtools'
The following objects are masked from 'package:remotes':
dev_package_deps, install_bioc, install_bitbucket, install_cran,
install_deps, install_dev, install_git, install_github,
install_gitlab, install_local, install_svn, install_url,
install_version, update_packages
library(png)
Warning: package 'png' was built under R version 4.3.3
library(MRPRESSO)library(vcfR)
***** *** vcfR *** *****
This is vcfR 1.15.0
browseVignettes('vcfR') # Documentation
citation('vcfR') # Citation
***** ***** ***** *****
library(data.table)
Warning: package 'data.table' was built under R version 4.3.3
Attaching package: 'data.table'
The following objects are masked from 'package:lubridate':
hour, isoweek, mday, minute, month, quarter, second, wday, week,
yday, year
The following objects are masked from 'package:dplyr':
between, first, last
The following object is masked from 'package:purrr':
transpose
An OpenGWAS token API was loaded to allow access to GWAS summary statistics from OpenGWAS database.
#Sys.getenv("OPENGWAS_JWT") #prevent it from appearing in the html fileoptions(ieugwasr.cache =TRUE)
4 Age at Menarche
4.1 Age at Menarche and Type 2 Diabetes
4.1.1Forward MR
Direction: AAM → T2D Exposure: Age at Menarche Outcome: Type 2 Diabetes Research Question: Does age at menarche causally influence T2D risk?
#################################################### Forward MR: AAM -> T2D###################################################obtain the AAM exposure dataset from openGWASexposure_aam <-extract_instruments("ieu-a-1095") #AAM#clumpingclumped_exp_aam <-clump_data(exposure_aam,clump_r2=0.001,clump_kb=10000, pop="EUR")
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping ieu-a-1095, 68 variants, using EUR population reference
Removing 2 of 68 variants due to LD with other variants or absence from LD reference panel
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
out_aam_T2D
id.exposure id.outcome exposure
1 ieu-a-1095 ebi-a-GCST005047 Age at menarche || id:ieu-a-1095
outcome snp_r2.exposure snp_r2.outcome
1 Type 2 diabetes || id:ebi-a-GCST005047 0.02548295 0.01417032
correct_causal_direction steiger_pval
1 TRUE 9.771126e-06
4.1.2 Results for Forward MR (AAM → T2D)
Method
β (Effect)
SE
p-value
IVW
-0.039
0.0631
0.5371
Not significant
Weighted median
0
0.0664
1
Not significant
MR-Egger
0.4715
0.2257
0.0407
Significant
Simple Mode
0.2149
0.1659
0.1720
Not significant
Weighted mode
0.2236
0.1312
0.0803
Not significant
MR scatter plot: The SNP points are widely scattered with most of the estimates lines lying close to 0 except for the MR-Egger line, which shows a noticeable positive slope. Since most of the MR estimates show no causal effect, the MR-Egger is likely influenced by pleiotropy.
Sensitivity analyses
Heterogeneity -> Strong heterogeneity was observed, indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> the egger intercept was significant too, indicating directional pleiotropy is present, violating MR assumption of no pleiotropy.
Steiger directionality test -> p = 9.77 × 10⁻⁶, confirming the correct causal direction is AAM to T2D, and that reverse causation is unlikely.
Single SNP analysis( and the forest plot) -> Almost all the SNPs were non-significant, showing no strong outliers dominating the results.
Funnel plot: The SNPs show asymmetrical distribution. The MR Egger and IVW do not overlap indicating directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are close to 0 with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: age at menarche does not causally influence T2D risk.
4.2Forward Multivariable MR (MVMR)
Direction: AAM + BMI → T2D
Exposures: Age at Menarche, Body Mass Index
Outcome: Type 2 Diabetes
Research Question: Does age at menarche affect T2D independent of BMI? and how does BMI affect T2D independent of AAM.
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping 1, 145 variants, using EUR population reference
Removing 24 of 145 variants due to LD with other variants or absence from LD reference panel
Extracting data for 121 SNP(s) from 2 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 4
Querying variant chunk 2 of 4
Querying variant chunk 3 of 4
Querying variant chunk 4 of 4
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1518080, rs1558902, rs1874984, rs4242496, rs4801589, rs9373571
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Type 2 diabetes || id:ebi-a-GCST005047 (ebi-a-GCST005047)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs14810, rs17001654, rs2617056, rs2836950, rs3914188, rs9304665, rs9555810, rs9579083, rs9914578
$result
id.exposure exposure id.outcome
1 ieu-a-1095 Age at menarche || id:ieu-a-1095 ebi-a-GCST005047
2 ieu-a-2 Body mass index || id:ieu-a-2 ebi-a-GCST005047
outcome nsnp b se pval
1 Type 2 diabetes || id:ebi-a-GCST005047 49 0.1014863 0.1084571 0.34941378
2 Type 2 diabetes || id:ebi-a-GCST005047 59 0.5860446 0.1858535 0.00161457
# Main MVMR plotggplot(res_mvmr$result, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.1 ) +theme_bw() +labs(title ="Multivariable MR: Direct effects on Type 2 Diabetes",x ="Exposure",y ="Effect on T2D" )
4.2.1Results for Forward Multivariable MR (AAM + BMI→ T2D)
Exposure
β
SE
p-value
AAM
0.1014
0.1084
0.3494
BMI
0.5860
0.1858
0.0016
Looking at the MVMR results, age at menarche had insignificant p-values (0.34941378) while BMI had significant p-values (0.00161457). After adjusting for BMI, age at menarche has no direct causal effects on T2D, confirming our forward MR conclusion. However, BMI has strong direct causal effect on type 2 diabetes.
4.3Reverse MR
Direction: T2D → AAM
Exposure: Type 2 Diabetes
Outcome: Age at Menarche
Research Question: Does T2D causally influence age at menarche?
####################################################Reverse MR: T2D -> AAM###################################################Instruments for T2D (exposure)exp_T2D <-extract_instruments("ebi-a-GCST005047") #t2d#ld clumpingexp_T2D_clumped <-clump_data(exp_T2D, clump_r2 =0.001, clump_kb =10000, pop ="EUR")
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping ebi-a-GCST005047, 34 variants, using EUR population reference
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
out_T2D_aam
id.exposure id.outcome exposure
1 ebi-a-GCST005047 ieu-a-1095 Type 2 diabetes || id:ebi-a-GCST005047
outcome snp_r2.exposure snp_r2.outcome
1 Age at menarche || id:ieu-a-1095 0.1541034 0.000592452
correct_causal_direction steiger_pval
1 TRUE 0
4.3.1 Results for Reverse MR (T2D → AAM)
Method
β
SE
p-value
IVW
−0.0117
0.0187
0.5324
Not significant
Weighted median
0.0095
0.0170
0.5762
Not significant
MR-Egger
0.0840
0.0467
0.8172
Not significant
Simple mode
-0.0213
0.0373
0.5268
Not significant
Weighted mode
0.0250
0.0202
0.2234
Not significant
MR scatter plot: The SNP points are mostly distributed close to 0 and the MR estimates points show a similar pattern except for the MR-Egger line, which shows a noticeable positive slope. Since most of the MR estimates show no causal effect, the MR-Egger is likely influenced by pleiotropy.
Sensitivity analyses
Heterogeneity -> Strong heterogeneity was observed( MR Egger and IVW Qp-values 8.245362e-08 and 1.029073e-09 respectively), indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> The egger intercept was significant, indicating directional pleiotropy is present, violating MR assumption of no pleiotropy, meaning some T2D affects AAM through pathways other than T2D.
Steiger directionality test -> p = 0 which is significant, causal direction is correct T2D to AAM.
Single SNP analysis (and the forest plots)-> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with no consistent patterns. Even though MR-Egger is slightly positive and IVW estimate, they both overlap 0, thus not significant, indicating no causal effect of type 2 diabetes on age at menarche.
Funnel plot: The SNPs show asymmetrical distribution, with most SNPs below 0 indicating directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are less than 0 and near -0.01 with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: This MR analysis provides no evidence that the T2D has a direct causal effect on age at menarche.
4.4Reverse Multivariable MR (MVMR)
Direction: T2D + BMI → AAM
Exposures: Type 2 Diabetes, Body Mass Index
Outcome: Age at Menarche
Research Question: Does T2D influence age at menarche independent of BMI? and does BMI causally influence T2D independent of AAM.
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping 1, 112 variants, using EUR population reference
Removing 11 of 112 variants due to LD with other variants or absence from LD reference panel
Extracting data for 101 SNP(s) from 2 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 4
Querying variant chunk 2 of 4
Querying variant chunk 3 of 4
Querying variant chunk 4 of 4
Finding proxies for 1 SNPs in outcome ebi-a-GCST005047
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Type 2 diabetes || id:ebi-a-GCST005047 (ebi-a-GCST005047) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs10258074, rs14810, rs1558902, rs17001654, rs4237150, rs6494307, rs9304665, rs9579083, rs9914578
id.exposure exposure id.outcome
1 ebi-a-GCST005047 Type 2 diabetes || id:ebi-a-GCST005047 ieu-a-1095
2 ieu-a-2 Body mass index || id:ieu-a-2 ieu-a-1095
outcome nsnp b se pval
1 Age at menarche || id:ieu-a-1095 26 -0.02802139 0.01820380 1.237274e-01
2 Age at menarche || id:ieu-a-1095 67 -0.53030869 0.05138305 5.681115e-25
#Plot reverse MVMRggplot(res_mvmr_rev_T2D_aam$result, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.1 ) +theme_bw() +labs(title ="Reverse MVMR: Effects on Age at menarche", x ="Exposure",y ="Effect on AAM" )
4.4.1Results for Reverse Multivariable MR (T2D + BMI → AAM)
Exposure
β
SE
p-value
T2D
-0.028
0.018
1.237274e-01
BMI
-0.530
0.051
5.681115e-25
The p-values for T2D was insignificant, indicating that T2D has no direct causal effect on age at menarche even after adjusting for BMI. However, BMI was significant with a negative beta, suggesting that higher BMI leads to earlier age at menarche independent of T2D.
This suggests that BMI rather than T2D is a primary causal determinant for early puberty.
4.5 Age at Menarche and Lipid levels
Lipids used LDL, HDL, Triglycerides (TG) and Total Cholesterol (TC)
4.6Forward MR
Direction: AAM → LDL, AAM → HDL, AAM → TG, AAM →TC
Exposure: Age at Menarche
Outcome: LDL, HDL, TG, TC
Research Question: Does age at menarche causally influence each lipid levels?
Finding proxies for 1 SNPs in outcome ebi-a-GCST002222
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs11756454, rs1518080, rs17351680, rs1874984, rs2617056, rs2836950, rs3914188, rs4242496, rs4801589, rs9373571, rs9555810, rs9939609
Analysing 'ieu-a-1095' on 'ebi-a-GCST002222'
Extracting data for 66 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and HDL cholesterol || id:ebi-a-GCST002223 (ebi-a-GCST002223)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs11756454, rs1518080, rs17351680, rs1874984, rs2617056, rs2836950, rs3914188, rs4242496, rs4801589, rs9373571, rs9555810, rs9939609
Analysing 'ieu-a-1095' on 'ebi-a-GCST002223'
Extracting data for 66 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Triglycerides || id:ebi-a-GCST002216 (ebi-a-GCST002216)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs11756454, rs1518080, rs17351680, rs1874984, rs2617056, rs2836950, rs3914188, rs4242496, rs4801589, rs9373571, rs9555810, rs9939609
Analysing 'ieu-a-1095' on 'ebi-a-GCST002216'
Extracting data for 66 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002221
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Cholesterol, total || id:ebi-a-GCST002221 (ebi-a-GCST002221)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs11756454, rs1518080, rs17351680, rs1874984, rs2617056, rs2836950, rs3914188, rs4242496, rs4801589, rs9373571, rs9555810, rs9939609
Analysing 'ieu-a-1095' on 'ebi-a-GCST002221'
mr_results_aam_to_lipids
$`ebi-a-GCST002222`
id.exposure id.outcome outcome
1 ieu-a-1095 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222
2 ieu-a-1095 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222
3 ieu-a-1095 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222
4 ieu-a-1095 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222
5 ieu-a-1095 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222
exposure method nsnp b
1 Age at menarche || id:ieu-a-1095 MR Egger 53 -0.027240702
2 Age at menarche || id:ieu-a-1095 Weighted median 53 0.008413812
3 Age at menarche || id:ieu-a-1095 Inverse variance weighted 53 0.006126733
4 Age at menarche || id:ieu-a-1095 Simple mode 53 -0.009294822
5 Age at menarche || id:ieu-a-1095 Weighted mode 53 -0.002438460
se pval
1 0.08054616 0.7366009
2 0.02471717 0.7335526
3 0.02080083 0.7683432
4 0.05717451 0.8714876
5 0.04188517 0.9537983
$`ebi-a-GCST002223`
id.exposure id.outcome outcome
1 ieu-a-1095 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223
2 ieu-a-1095 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223
3 ieu-a-1095 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223
4 ieu-a-1095 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223
5 ieu-a-1095 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223
exposure method nsnp b
1 Age at menarche || id:ieu-a-1095 MR Egger 53 0.11165469
2 Age at menarche || id:ieu-a-1095 Weighted median 53 0.07710385
3 Age at menarche || id:ieu-a-1095 Inverse variance weighted 53 0.07230400
4 Age at menarche || id:ieu-a-1095 Simple mode 53 0.03836100
5 Age at menarche || id:ieu-a-1095 Weighted mode 53 0.05195367
se pval
1 0.07601272 0.1480029741
2 0.02474681 0.0018350357
3 0.01973294 0.0002481833
4 0.05916391 0.5195889779
5 0.05705178 0.3666868012
$`ebi-a-GCST002216`
id.exposure id.outcome outcome
1 ieu-a-1095 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216
2 ieu-a-1095 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216
3 ieu-a-1095 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216
4 ieu-a-1095 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216
5 ieu-a-1095 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216
exposure method nsnp b
1 Age at menarche || id:ieu-a-1095 MR Egger 53 0.007630288
2 Age at menarche || id:ieu-a-1095 Weighted median 53 -0.042285086
3 Age at menarche || id:ieu-a-1095 Inverse variance weighted 53 -0.037800973
4 Age at menarche || id:ieu-a-1095 Simple mode 53 -0.034029896
5 Age at menarche || id:ieu-a-1095 Weighted mode 53 -0.030079878
se pval
1 0.07662058 0.92106395
2 0.02408466 0.07914211
3 0.01988395 0.05729160
4 0.04934471 0.49348927
5 0.03504979 0.39471822
$`ebi-a-GCST002221`
id.exposure id.outcome outcome
1 ieu-a-1095 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221
2 ieu-a-1095 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221
3 ieu-a-1095 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221
4 ieu-a-1095 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221
5 ieu-a-1095 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221
exposure method nsnp b
1 Age at menarche || id:ieu-a-1095 MR Egger 53 2.914114e-02
2 Age at menarche || id:ieu-a-1095 Weighted median 53 1.878874e-03
3 Age at menarche || id:ieu-a-1095 Inverse variance weighted 53 1.311659e-02
4 Age at menarche || id:ieu-a-1095 Simple mode 53 -1.998854e-02
5 Age at menarche || id:ieu-a-1095 Weighted mode 53 2.152908e-05
se pval
1 0.08542199 0.7343972
2 0.02522065 0.9406146
3 0.02211328 0.5530778
4 0.05310002 0.7081275
5 0.04414909 0.9996128
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
out_aam_to_tc
id.exposure id.outcome exposure
1 ieu-a-1095 ebi-a-GCST002221 Age at menarche || id:ieu-a-1095
outcome snp_r2.exposure snp_r2.outcome
1 Cholesterol, total || id:ebi-a-GCST002221 0.02548295 0.001376164
correct_causal_direction steiger_pval
1 TRUE 5.979091e-210
4.6.1 Results for Forward MR (AAM → Lipids)
4.6.2 LDL cholesterol (AAM → LDL)
Method
β
SE
p-value
IVW
0.0061
0.0208
0.7683
Not significant
Weighted median
0.0084
0.0260
0.7420
Not significant
MR-Egger
-0.0272
0.0805
0.7366
Not significant
Simple mode
-0.0092
0.0530
0.8614
Not significant
Weighted mode
-0.0024
0.0450
0.9542
Not significant
As seen on the table above and MR scatter plot, the estimates lie close to 0 and are insignificant, suggesting no causal effect of AAM to LDL levels.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 0.0047 and 0.0039 respectively), indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.67), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 3.503484e-224, significant, causal direction is correct, AAM to LDL.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap but also cross 0, thus not significant, indicating no causal effect of age at menarche on LDL.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are more than 0 except one with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: The MR analysis provides no evidence that age at menarche has a causal effect on LDL cholesterol.
4.6.2.1 HDL cholesterol (AAM → HDL)
Method
β
SE
p-value
IVW
0.0723
0.0197
0.0002
Significant
Weighted median
0.0771
0.0251
0.0017
Significant
MR-Egger
0.1116
0.0760
0.1480
Not significant
Simple mode
0.0383
0.0588
0.5169
Not significant
Weighted mode
0.0519
0.0541
0.3417
Not significant
As seen on the table above and MR scatter plot, the beta estimates suggest that later AAM increases HDL cholesterol but all the estimates are negative except for IVW and Weighted Median. The significance suggest that higher later age of menarche is associated with higher HDL.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed (MR Egger and IVW Qp-values 0.001 and 0.0012 respectively), indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.594), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 6.19622e-207, significant, causal direction is correct, AAM to HDL.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap, with MR-Egger crossing 0.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are more than 0 except one with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: Age at menarche has a positive causal effect on HDL cholesterol.
4.6.2.2 Triglycerides (AAM →TG)
Method
β
SE
p-value
IVW
-0.0378
0.0199
0.0573
Not significant
Weighted median
-0.0422
0.0231
0.0581
Not significant
MR-Egger
0.0076
0.0766
0.9210
Not significant
Simple mode
-0.0340
0.0494
0.4943
Not significant
Weighted mode
-0.0300
0.0382
0.4346
Not significant
The MR estimates from the table above are show that age of menarche has no causal effect on TG levels. The scatter plot shows no clear linear relationship between SNP effects on age at menarche and TG levels.
Heterogeneity -> Significant heterogeneity was observed (MR Egger and IVW Qp-values 0.00027 and 0.00031 respectively), indicating inconsistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.542 ), indicating no evidence of directional horizontal pleiotropy.
Steiger directionality test -> p = 3.8 × 10⁻²¹², significant, causal direction is correct, AAM to TG.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap, with MR-Egger crossing 0.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are more than 0 except one with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: The MR analyses show no direct causal relationship of age at menarche on TG levels.
4.6.2.3 Total Cholesterol (AAM → TC)
Method
β
SE
p-value
IVW
1.311659e-02
0.0221
0.5531
Not significant
Weighted median
1.878874e-03
0.0259
0.9422
Not significant
MR-Egger
2.914114e-02
0.0854
0.7343
Not significant
Simple mode
-1.998854e-02
0.0518
0.7014
Not significant
Weighted mode
2.152908e-05
0.0421
0.9996
Not significant
The MR estimates from the table above are show that age of menarche has no causal effect on Tc levels. The scatter plot shows no clear linear relationship between SNP effects on age at menarche and TClevels.
Heterogeneity -> Significant heterogeneity was observed (MR Egger and IVW Qp-values 6.2 × 10⁻⁵and 8.7 × 10⁻⁵ respectively), indicating inconsistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.8467), indicating no evidence of directional horizontal pleiotropy.
Steiger directionality test -> p = 6 × 10⁻²¹⁰, significant, causal direction is correct, AAM to TC.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap and cross 0.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are more than 0 except one with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: The MR analyses show no direct causal relationship of age at menarche on TC levels.
4.6.3Forward Multivariable MR (MVMR)
Direction: AAM + BMI → HDL or LDL or TG or TC)
Exposures: Age at Menarche, Body Mass Index
Outcome: HDL, LDL, TG and TC
Research Question: Does age at menarche influence the each lipid level independent of BMI? and does BMI affect the AMM independent of each lipid level.
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping 1, 145 variants, using EUR population reference
Removing 24 of 145 variants due to LD with other variants or absence from LD reference panel
Extracting data for 121 SNP(s) from 2 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 4
Querying variant chunk 2 of 4
Querying variant chunk 3 of 4
Querying variant chunk 4 of 4
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1518080, rs1558902, rs1874984, rs4242496, rs4801589, rs9373571
Finding proxies for 1 SNPs in outcome ebi-a-GCST002222
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs14810, rs17001654, rs2617056, rs2836950, rs3914188, rs9304665, rs9555810, rs9579083, rs9914578
Extracting data for 115 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002223
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and HDL cholesterol || id:ebi-a-GCST002223 (ebi-a-GCST002223)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs14810, rs17001654, rs2617056, rs2836950, rs3914188, rs9304665, rs9555810, rs9579083, rs9914578
Extracting data for 115 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002216
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Triglycerides || id:ebi-a-GCST002216 (ebi-a-GCST002216)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs14810, rs17001654, rs2617056, rs2836950, rs3914188, rs9304665, rs9555810, rs9579083, rs9914578
Extracting data for 115 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 2 SNPs in outcome ebi-a-GCST002221
Extracting data for 2 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menarche || id:ieu-a-1095 (ieu-a-1095) and Cholesterol, total || id:ebi-a-GCST002221 (ebi-a-GCST002221)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs1079866, rs14810, rs17001654, rs2617056, rs2836950, rs3914188, rs9304665, rs9555810, rs9579083, rs9914578
# Plot MVMR for all lipidslipid_names <-c("ebi-a-GCST002222"="LDL Cholesterol","ebi-a-GCST002223"="HDL Cholesterol","ebi-a-GCST002216"="Triglycerides","ebi-a-GCST002221"="Total Cholesterol")for(lipid in lipids){ res_obj <- mvmr_results_bmi_aam_to_lipids[[lipid]]# Skip if result is missing, NA, or malformedif (is.null(res_obj) ||is.na(res_obj)[1] ||!is.list(res_obj) ||!"result"%in%names(res_obj)) {message(paste("Skipping", lipid, "- no valid MVMR result"))next } res_mvmr <- res_obj$result p <-ggplot(res_mvmr, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.1 ) +theme_bw() +labs(title =paste("MVMR: Direct effects of AAM and BMI on", lipid_names[lipid]),x ="Exposure",y ="Effect on Lipid" )print(p)}
Skipping ebi-a-GCST002222 - no valid MVMR result
Skipping ebi-a-GCST002223 - no valid MVMR result
Skipping ebi-a-GCST002216 - no valid MVMR result
Skipping ebi-a-GCST002221 - no valid MVMR result
nrow(out_mvmr_bmi_aam_to_lipid)
[1] 114
nrow(mv_dat_bmi_aam_to_lipid)
NULL
4.6.3.1Results for Multivariable MR (AAM +BMI → HDL or LDL or TG or TC)
Although the outcome data was extracted for all the lipid traits, MV harmonisation failed because no SNPs had valid association estimates for the three AAM, BMI and lipid outcome. Therefore, MVMR could not be performed due to insufficient lipid overlaps.
4.7 Reverse MR
Direction: Lipids → AAM Exposures: Lipid Traits, Body Mass Index Outcome: Age at Menarche Research Question: Do lipid levels influence age at menarche independent of BMI? and does BMI influence AAM independent of each lipid level.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
out_tc_to_aam
id.exposure id.outcome exposure
1 ebi-a-GCST002221 ieu-a-1095 Cholesterol, total || id:ebi-a-GCST002221
outcome snp_r2.exposure snp_r2.outcome
1 Age at menarche || id:ieu-a-1095 0.1035706 0.0005798003
correct_causal_direction steiger_pval
1 TRUE 0
4.7.1 Results for Reverse MR (Lipids → AAM)
4.7.1.1 LDL cholesterol (LDL → AAM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
0.0403
0.0330
0.2275
Not significant
Weighted median
0.0020
0.0275
0.9416
Not significant
IVW
-0.0120
0.0185
0.5146
Not significant
Simple mode
-0.0132
0.0491
0.7881
Not significant
Weighted mode
0.0019
0.0254
0.9410
Not significant
As seen on the table above and MR scatter plot, the estimates lie close to 0 and are insignificant, suggesting no causal effect of LDL levels on AAM .
Sensitivity analyses
Heterogeneity -> Insignificant heterogeneity was observed ( MR Egger and IVW Qp-values 0.1459 and 0.2080 respectively), indicating consistent SNP effects and suggesting no possible pleiotropic effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.67), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, LDL to AAM.
Single SNP analysis (and the forest plot) -> The SNPs show insignificant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions. MR-Egger and IVW summary estimates overlap but also cross 0, thus not significant, indicating no specific outliers.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
Conclusion: LDL levels have no causal effect on AAM.
4.7.1.2 HDL cholesterol (HDL → AAM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
-0.0102
0.0279
0.7159
Not significant
Weighted median
-0.0183
0.0246
0.4559
Not significant
IVW
0.0015
0.0170
0.9287
Not significant
Simple mode
0.0545
0.0472
0.2518
Not significant
Weighted mode
-0.0090
0.0212
0.6722
Not significant
As seen on the table above and MR scatter plot, the estimates lie close to 0 with some showing negative and positive effect sizes but the P-values are insignificant, suggesting no causal effect of HDL levels on AAM.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 0.0315and 0.0352 respectively), indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.5977), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, HDL to AAM.
Single SNP analysis (and the forest plots) -> The SNPs show insignificant (p-values >0.05 and CI crossing 0). MR-Egger and IVW summary estimates overlap and close to 0, thus not significant, indicating no specific outliers.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
Conclusion: HDL levels have no causal effect on AAM.
4.7.1.3 Triglycerides (TG → AAM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
0.0845
0.0393
0.0360
Significant
Weighted median
0.0414
0.0318
0.1930
Not significant
IVW
0.0169
0.0247
0.4942
Not significant
Simple mode
-0.0259
0.0537
0.6322
Not significant
Weighted mode
0.0297
0.0301
0.3291
Not significant
As seen on the table above and MR scatter plot, the estimates lie close to 0 with some most showing positive effect sizes but the p-values are insignificant, suggesting no causal effect of TG levels on AAM.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 0.0029 and 0.0007 respectively), indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> The MR egger intercept was significant (p-value 0.0347), indicating directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, TG to AAM.
Single SNP analysis (and the forest plots) -> The SNPs show insignificant (p-values >0.05 and CI crossing 0). MR-Egger and IVW summary estimates overlap and close to 0, thus not significant, indicating no specific outliers.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
Conclusion: TG levels have no causal effect on AAM.
4.7.1.4Total cholesterol (TC → AAM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
0.0548
0.0367
0.1390
Not significant
Weighted median
0.0287
0.0260
0.2694
Not significant
IVW
0.0089
0.0184
0.6298
Not significant
Simple mode
0.0148
0.0511
0.7734
Not significant
Weighted mode
0.0241
0.0308
0.4353
Not significant
As seen on the table above and MR scatter plot, the estimates lie close to 0 with some most showing positive effect sizes but the p-values are insignificant, suggesting no causal effect of TC levels on AAM.
Sensitivity analyses
Heterogeneity -> Borderline significant heterogeneity was observed ( MR Egger and IVW Qp-values 0.0512 and 0.0413 respectively), indicating some SNOs inconsistency.
MR-Egger pleiotropy test -> The MR egger intercept was insignificant (p-value 0.1523), indicating no directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, TC to AAM.
Single SNP analysis (and the forest plots) -> The SNPs show insignificant (p-values >0.05 and CI crossing 0). MR-Egger and IVW summary estimates overlap and close to 0, thus not significant, indicating no specific outliers.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
Conclusion: TC levels have no causal effect on AAM.
4.7.2Reverse Multivariable MR (MVMR)
Direction: Lipids + BMI → AAM
Exposures: HDL, LDL, TG and TC , Body Mass Index
Outcome: Age at Menarche
Research Question: Do lipid levels influence age at menarche independent of BMI? and does BMI affect the each lipid level independent of AAM.
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs7534572, rs9491696, rs9914578
Harmonising LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222) and Cholesterol, total || id:ebi-a-GCST002221 (ebi-a-GCST002221)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs7534572, rs9491696, rs9914578
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs7534572, rs9491696, rs9914578
Harmonising LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs9491696, rs9914578
Harmonising LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222) and Age at menarche || id:ieu-a-1095 (ieu-a-1095)
# 4. Run multivariable MRres_mvmr_lipids_to_aam_bmi <-mv_multiple(mvmr_dat_lipids_to_aam_bmi)res_mvmr_lipids_to_aam_bmi$result
id.exposure exposure id.outcome
1 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216 ieu-a-1095
2 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221 ieu-a-1095
3 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222 ieu-a-1095
4 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223 ieu-a-1095
5 ieu-a-2 Body mass index || id:ieu-a-2 ieu-a-1095
outcome nsnp b se pval
1 Age at menarche || id:ieu-a-1095 35 0.01878774 0.10761983 8.614135e-01
2 Age at menarche || id:ieu-a-1095 67 -0.05392406 0.29254452 8.537564e-01
3 Age at menarche || id:ieu-a-1095 53 0.02567207 0.25491217 9.197811e-01
4 Age at menarche || id:ieu-a-1095 69 0.01705047 0.12865837 8.945690e-01
5 Age at menarche || id:ieu-a-1095 51 -0.54435966 0.04772129 3.855368e-30
#Plot reverse MVMRggplot(res_mvmr_lipids_to_aam_bmi$result, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.15 ) +theme_bw() +theme(axis.text.x =element_text(angle =30, hjust =1, size =11) ) +labs(title ="Reverse MVMR: Effects of Lipids and BMI on Age at Menarche",x ="Exposure",y ="Effect on AAM" )
4.7.2.1 Results for Reverse MVMR (Lipids + BMI → AAM)
Exposure
SNPs
β
SE
p-value
Triglycerides
35
0.0188
0.1076
0.8614
Not significant
Total Cholesterol
67
-0.0539
0.2925
0.8538
Not significant
LDL Cholesterol
53
0.0257
0.2549
0.9198
Not significant
HDL Cholesterol
69
0.0171
0.1287
0.8946
Not significant
Body Mass Index (BMI)
51
-0.5444
0.0477
3.86 × 10⁻³⁰
Highly significant
Only BMI showed significant causal effects on AAM, suggesting higher BMI leads to earlier age at menarche. Higher BMI, early puberty.
5Age at Natural Menopause
5.1 Age at natural menopause and Type 2 Diabetes
5.2 Forward MR
Direction: ANM → T2D Exposure: ANM Outcome: Type 2 Diabetes Research Question: Does ANM causally influence T2D risk?
#################################################### Forward MR: ANM -> T2D################################################### Obtain the ANM exposure dataset from OpenGWASexposure_anm <-extract_instruments("ieu-a-1004") #age at natural menopause# Clumpingclumped_exp_anm <-clump_data( exposure_anm,clump_r2 =0.001,clump_kb =10000,pop ="EUR")
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping ieu-a-1004, 43 variants, using EUR population reference
Removing 2 of 43 variants due to LD with other variants or absence from LD reference panel
### F statisticclumped_exp_anm$F_statistic <- (clumped_exp_anm$beta.exposure^2) / (clumped_exp_anm$se.exposure^2)summary(clumped_exp_anm$F_statistic)
Min. 1st Qu. Median Mean 3rd Qu. Max.
28.44 42.25 56.25 80.95 81.00 484.00
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
head(out_anm_T2D)
id.exposure id.outcome exposure
1 ieu-a-1004 ebi-a-GCST005047 Age at menopause || id:ieu-a-1004
outcome snp_r2.exposure snp_r2.outcome
1 Type 2 diabetes || id:ebi-a-GCST005047 0.04007178 0.007322349
correct_causal_direction steiger_pval
1 TRUE 9.514891e-33
5.2.1Results for Forward MR (ANM → T2D)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
-0.0827
0.0478
0.0916
Not significant
Weighted Median
-0.0471
0.0206
0.0219
Significant
IVW
-0.0183
0.0188
0.3307
Not significant
Simple Mode
-0.0624
0.0401
0.1256
Not significant
Weighted Mode
-0.0624
0.0225
0.0084
Significant
The MR estimates were non significant except for Weighted Median and Weighted Mode.
Sensitivity analyses
Heterogeneity -> Strong heterogeneity was observed, indicating inconsistent SNP effects and suggesting possible pleiotropic effects (MR-Egger p-value 2.779106e-05 and IVW Qpvalue 1.086847e-05).
MR-Egger pleiotropy test -> The Egger intercept was not significant too (p-value 0.1518), indicating no strong evidence of directional pleiotropy.
Steiger directionality test -> p = 9.514891e-33, confirming the correct causal direction is ANM to T2D, and that reverse causation is unlikely.
Single SNP analysis( and the forest plot) -> Almost all the SNPs were non-significant, with some having positive and negative effect but showed no strong outliers dominating the results.
Funnel plot: The SNPs show symmetrical distribution.
Leave-one-out analysis -> Most p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. Most of the estimates had a confidence interval crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: Suggestive evidence that later age of menopause may be associated with reduced T2D. In the presence of strong heterogeneity as observed, this should be interpreted cautiously.
5.3 Multivariable MR
5.4Forward Multivariable MR (MVMR)
Direction: ANM + BMI → T2D
Exposures: ANM, Body Mass Index
Outcome: Type 2 Diabetes
Research Question: Does ANM affect T2D independent of BMI? and how does BMI affect T2D independent of ANM?
Harmonising Age at menopause || id:ieu-a-1004 (ieu-a-1004) and Type 2 diabetes || id:ebi-a-GCST005047 (ebi-a-GCST005047)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs11031002, rs11852419, rs14810, rs1727326, rs9304665, rs9579083
$result
id.exposure exposure id.outcome
1 ieu-a-1004 Age at menopause || id:ieu-a-1004 ebi-a-GCST005047
2 ieu-a-2 Body mass index || id:ieu-a-2 ebi-a-GCST005047
outcome nsnp b se
1 Type 2 diabetes || id:ebi-a-GCST005047 32 -0.009090498 0.03196364
2 Type 2 diabetes || id:ebi-a-GCST005047 66 0.599057008 0.16617247
pval
1 0.7761029629
2 0.0003121151
# Main MVMR plotggplot(res_mvmr_anm$result, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.1 ) +theme_bw() +labs(title ="Multivariable MR: Direct effects on Type 2 Diabetes",x ="Exposure",y ="Effect on T2D" )
After adjusting for BMI, age at natural menopause has no direct causal effect on type 2 diabetes. However, Higher BMI has a strong direct causal effect increasing the risk of T2D.
5.5Reverse MR
Direction: T2D → ANM
Exposure: Type 2 Diabetes
Outcome: ANM
Research Question: Does T2D causally influence ANM?
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
head(out_T2D_anm)
id.exposure id.outcome exposure
1 ebi-a-GCST005047 ieu-a-1004 Type 2 diabetes || id:ebi-a-GCST005047
outcome snp_r2.exposure snp_r2.outcome
1 Age at menopause || id:ieu-a-1004 0.1541034 0.00037895
correct_causal_direction steiger_pval
1 TRUE 0
5.5.0.1 Results for Reverse MR (T2D → ANM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
0.0988
0.0916
0.2885
Not significant
Weighted Median
0.0470
0.0557
0.3988
Not significant
IVW
0.0512
0.0347
0.1403
Not significant
Simple Mode
0.0586
0.0915
0.5262
Not significant
Weighted Mode
0.0490
0.0621
0.4354
Not significant
Based on the effects on the table and their distribution on the MR scatter plot, T2D could delay menopause, but since all of the estimates are not significant, therefore no causal effect of T2D on ANM
Sensitivity analyses
Heterogeneity -> Heterogeneity was not observed (MR Egger and IVW Qp-values 0.5590 and 0.5925 respectively), indicating consistent SNP effects and suggesting no possible pleiotropic effects.
MR-Egger pleiotropy test -> The egger intercept was not significant (p-value 0.5780), indicating no directional pleiotropy.
Steiger directionality test -> p = 0 which is significant, causal direction is correct T2D to ANM.
Single SNP analysis (and the forest plots) -> Most of the SNPs were not significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with no consistent patterns. The MR-Egger and IVW estimate, were overlapping and crossing 0, thus not significant, indicating no single SNP showed significant drivers.
Funnel plot: The SNPs show symmetrical distribution.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates were more than 0 but the confidence intervals were crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: This MR analysis provides no evidence that the T2D has a direct causal effect on age at natural menopause.
5.6Reverse Multivariable MR (MVMR)
Direction: T2D + BMI → ANM
Exposures: Type 2 Diabetes, Body Mass Index
Outcome: ANM
Research Question: Does T2D influence ANM independent of BMI? and does BMI causally influence T2D independent of ANM.
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping 1, 112 variants, using EUR population reference
Removing 11 of 112 variants due to LD with other variants or absence from LD reference panel
Extracting data for 101 SNP(s) from 2 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 4
Querying variant chunk 2 of 4
Querying variant chunk 3 of 4
Querying variant chunk 4 of 4
Finding proxies for 1 SNPs in outcome ebi-a-GCST005047
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Type 2 diabetes || id:ebi-a-GCST005047 (ebi-a-GCST005047) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs10258074, rs14810, rs1558902, rs17001654, rs4237150, rs6494307, rs9304665, rs9579083, rs9914578
# 2. Outcome = ANMout_anm_rev <-extract_outcome_data(snps = exp_mv_rev_anm$SNP,outcomes ="ieu-a-1004"# Age at Natural Menopause)
id.exposure exposure id.outcome
1 ebi-a-GCST005047 Type 2 diabetes || id:ebi-a-GCST005047 ieu-a-1004
2 ieu-a-2 Body mass index || id:ieu-a-2 ieu-a-1004
outcome nsnp b se pval
1 Age at menopause || id:ieu-a-1004 26 0.06515825 0.04620489 0.1584799
2 Age at menopause || id:ieu-a-1004 67 -0.20322877 0.13167932 0.1227431
# 5. Plot reverse MVMRggplot(res_mvmr_rev_T2D_anm$result, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.1 ) +theme_bw() +labs(title ="Reverse MVMR: Effects of T2D and BMI on Age at Natural Menopause",x ="Exposure",y ="Effect on ANM" )
Exposure
β
SE
p-value
T2D
0.0652
0.0462
0.1585
Not significant
Body Mass Index
-0.2032
0.1317
0.1227
Not significant
BMI and T2D does not influence age of natural menopause.
5.7 Age at Menopause (ANM) and Lipid levels
5.8Forward MR
Direction: ANM → LDL, ANM → HDL, ANM → TG, ANM →TC
Exposure: ANM
Outcome: LDL, HDL, TG, TC
Research Question: Does ANM causally influence each lipid levels?
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
out_anm_to_tc
id.exposure id.outcome exposure
1 ieu-a-1004 ebi-a-GCST002221 Age at menopause || id:ieu-a-1004
outcome snp_r2.exposure snp_r2.outcome
1 Cholesterol, total || id:ebi-a-GCST002221 0.04007178 0.0007322501
correct_causal_direction steiger_pval
1 TRUE 4.391643e-271
Based on the effects on the table and their distribution on the MR scatter plot, ANM could increase LDL levels, except for IVW estimates, but since all of the estimates are not significant, therefore no causal effect of ANM on LDL.
Sensitivity analyses
Heterogeneity -> Insignificant heterogeneity was observed ( MR Egger and IVW Qp-values 0.3250 and 0.3571 respectively), indicating consistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.6141), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 9.850347e-290, significant, causal direction is correct, ANM to LDL.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap but also cross 0, thus not significant, indicating no causal effect of ANM on LDL.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are less than 0 with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: The MR analysis provides no evidence that genetically predicted ANM has a causal effect on LDL levels.
5.8.1.2 HDL cholesterol (ANM → HDL)
Method
β (effect)
SE
p-value
Interpretation
MR-Egger
-0.0164
0.0105
0.1290
Not significant
Weighted median
-0.0114
0.0058
0.0481
Significant
IVW
-0.0043
0.0051
0.4063
Not significant
Simple mode
-0.0059
0.0127
0.6464
Not significant
Weighted mode
-0.0109
0.0059
0.0742
Not significant
As seen on the table above and MR scatter plot, the beta estimates suggest later ANM causes a decrease in HDL levels. The p-values are all non-significant except for weighted median.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed (MR Egger and IVW Qp-values 0.007305 and 0.00495 respectively), indicating inconsistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.1975), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 63.576071e-279, significant, causal direction is correct, ANM to HDL.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap, with MR-Egger crossing 0.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are less than 0 with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: Mendelian randomization analyses provided evidence that genetically predicted ANM has no causal effect on HDL cholesterol.
5.8.1.3 Triglycerides (ANM → TG)
Method
β (effect)
SE
p-value
Interpretation
MR-Egger
-0.0072
0.0273
0.7936
Not significant
Weighted median
0.0038
0.0054
0.4825
Not significant
IVW
0.0153
0.0132
0.2446
Not significant
Simple mode
0.0083
0.0090
0.3700
Not significant
Weighted mode
0.0047
0.0050
0.3966
Not significant
The MR estimates from the table above are show that age of menarche has no causal effect on TG levels. The scatter plot shows no clear linear relationship between SNP effects on AAM and TG levels.
Heterogeneity -> Significant heterogeneity was observed (MR Egger and IVW Qp-values 2.626883e-65 and 7.021756e-67 respectively), indicating consistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.3531), indicating no evidence of directional horizontal pleiotropy.
Steiger directionality test -> 1.412319e-165 significant, causal direction is correct, ANM to TG.
Single SNP analysis (and the forest plots) -> A few SNPs are significant but the majority are non significant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions and with most of them close to 0. MR-Egger and IVW summary estimates overlap, with MR-Egger crossing 0. However, one SNP is seen to be….
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the estimates are more than 0 with the confidence intervals crossing 0, indicating no SNP shifts/drives the effect.
Conclusion: The MR analyses show no direct causal relationship on ANM on TG levels.
5.8.1.4 Total cholesterol (ANM → TC)
Method
β (effect)
SE
p-value
Interpretation
MR-Egger
-0.0131
0.0119
0.2798
Not significant
Weighted median
-0.0050
0.0065
0.4424
Not significant
IVW
-0.0028
0.0058
0.6232
Not significant
Simple mode
-0.0015
0.0130
0.9055
Not significant
Weighted mode
-0.0072
0.0070
0.3108
Not significant
As seen on the table above and MR scatter plot, the estimates suggests ANM decreases total cholesterol levels but all the estimates are insignificant p>0.05.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 0.0005 and 0.0004 respectively), indicating SNP consistency.
MR-Egger pleiotropy test -> The MR egger intercept was insignificant (p-value 0.3322), indicating no directional pleiotropy.
Steiger directionality test -> p = 4.391643e-271, significant, causal direction is correct, ANM to TC.
Single SNP analysis (and the forest plots) -> Most of the SNPs show insignificant (p-values >0.05 and CI crossing 0), with a few showing significant values. MR-Egger and IVW summary estimates overlap and close to 0.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
5.8.2Forward Multivariable MR (MVMR)
Direction: AAM + BMI → HDL or LDL or TG or TC)
Exposures: Age at Menarche, Body Mass Index
Outcome: HDL, LDL, TG and TC
Research Question: Does age at menarche influence the each lipid level independent of BMI? and does BMI affect the AMM independent of each lipid level.
Please look at vignettes for options on running this locally if you need to run many instances of this command.
Clumping 1, 121 variants, using EUR population reference
Removing 13 of 121 variants due to LD with other variants or absence from LD reference panel
Extracting data for 108 SNP(s) from 2 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 4
Querying variant chunk 2 of 4
Querying variant chunk 3 of 4
Querying variant chunk 4 of 4
Harmonising Age at menopause || id:ieu-a-1004 (ieu-a-1004) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs11647580, rs12599106, rs1558902
# Storage objectsmvmr_results_bmi_anm_to_lipids <-list()mvmr_data_bmi_anm_to_lipids <-list()# 2. Loop over lipid outcomes: harmonise + run MVMR if possiblefor (lipid in lipids) {# Outcome = lipid out_mvmr_bmi_anm_to_lipid <-extract_outcome_data(snps = exp_mvmr_bmi_anm_to_lipids$SNP,outcomes = lipid )# Harmonise mv_dat_bmi_anm_to_lipid <-mv_harmonise_data(exposure_dat = exp_mvmr_bmi_anm_to_lipids,outcome_dat = out_mvmr_bmi_anm_to_lipid )# Store harmonised data safelyif (!is.null(mv_dat_bmi_anm_to_lipid) &&is.data.frame(mv_dat_bmi_anm_to_lipid)) { mvmr_data_bmi_anm_to_lipids[[lipid]] <- mv_dat_bmi_anm_to_lipid nsnps <-nrow(mv_dat_bmi_anm_to_lipid) } else { mvmr_data_bmi_anm_to_lipids[[lipid]] <-NA nsnps <-0 }cat("Lipid:", lipid, "| SNPs after harmonisation:", nsnps, "\n")# Only run MVMR if enough SNPsif (nsnps >3) { res_mvmr <-mv_multiple(mv_dat_bmi_anm_to_lipid) } else { res_mvmr <-NAcat("Skipping MVMR for", lipid, "- too few SNPs\n") }# Store results mvmr_results_bmi_anm_to_lipids[[lipid]] <- res_mvmr}
Extracting data for 105 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002222
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menopause || id:ieu-a-1004 (ieu-a-1004) and LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs11031002, rs11852419, rs14810, rs1727326, rs9304665, rs9579083
Lipid: ebi-a-GCST002222 | SNPs after harmonisation: 0
Skipping MVMR for ebi-a-GCST002222 - too few SNPs
Extracting data for 105 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002223
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menopause || id:ieu-a-1004 (ieu-a-1004) and HDL cholesterol || id:ebi-a-GCST002223 (ebi-a-GCST002223)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs11031002, rs11852419, rs14810, rs1727326, rs9304665, rs9579083
Lipid: ebi-a-GCST002223 | SNPs after harmonisation: 0
Skipping MVMR for ebi-a-GCST002223 - too few SNPs
Extracting data for 105 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002216
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menopause || id:ieu-a-1004 (ieu-a-1004) and Triglycerides || id:ebi-a-GCST002216 (ebi-a-GCST002216)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs11031002, rs11852419, rs14810, rs1727326, rs9304665, rs9579083
Lipid: ebi-a-GCST002216 | SNPs after harmonisation: 0
Skipping MVMR for ebi-a-GCST002216 - too few SNPs
Extracting data for 105 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 2
Querying variant chunk 2 of 2
Finding proxies for 1 SNPs in outcome ebi-a-GCST002221
Extracting data for 1 SNP(s) from 1 GWAS(s)
Querying id chunk 1 of 1
Querying variant chunk 1 of 1
Harmonising Age at menopause || id:ieu-a-1004 (ieu-a-1004) and Cholesterol, total || id:ebi-a-GCST002221 (ebi-a-GCST002221)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs11031002, rs11852419, rs14810, rs1727326, rs9304665, rs9579083
Lipid: ebi-a-GCST002221 | SNPs after harmonisation: 0
Skipping MVMR for ebi-a-GCST002221 - too few SNPs
# 3. Plot MVMR results (only for lipids where MVMR actually ran)lipid_names <-c("ebi-a-GCST002222"="LDL Cholesterol","ebi-a-GCST002223"="HDL Cholesterol","ebi-a-GCST002216"="Triglycerides","ebi-a-GCST002221"="Total Cholesterol")for (lipid in lipids) { res_obj <- mvmr_results_bmi_anm_to_lipids[[lipid]]# Skip if result is missing, NA, or malformedif (is.null(res_obj) ||is.na(res_obj)[1] ||!is.list(res_obj) ||!"result"%in%names(res_obj)) {message(paste(" Skipping plot for", lipid, "- no valid MVMR result"))next } res_mvmr <- res_obj$result p <-ggplot(res_mvmr, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.1 ) +theme_bw() +labs(title =paste("MVMR: Direct effects of ANM and BMI on", lipid_names[lipid]),x ="Exposure",y ="Effect on Lipid" )print(p)}
Skipping plot for ebi-a-GCST002222 - no valid MVMR result
Skipping plot for ebi-a-GCST002223 - no valid MVMR result
Skipping plot for ebi-a-GCST002216 - no valid MVMR result
Skipping plot for ebi-a-GCST002221 - no valid MVMR result
5.8.2.1Results for Forward MVMR (AAM + BMI → HDL or LDL or TG or TC)
Multivariable MR of ANM and BMI on lipid traits could not be performed because no SNPs remained after harmonization, indicating insufficient shared genetic instruments across exposures and outcomes
5.9 Reverse MR
Direction: Lipids → ANM Exposures: Lipid Traits, Body Mass Index Outcome: ANM Research Question: Do lipid levels influence ANM independent of BMI? and does BMI influence ANM independent of each lipid level.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
r.exposure and/or r.outcome not present.
Calculating approximate SNP-exposure and/or SNP-outcome correlations, assuming all are quantitative traits. Please pre-calculate r.exposure and/or r.outcome using get_r_from_lor() for any binary traits
out_tc_to_anm
id.exposure id.outcome exposure
1 ebi-a-GCST002221 ieu-a-1004 Cholesterol, total || id:ebi-a-GCST002221
outcome snp_r2.exposure snp_r2.outcome
1 Age at menopause || id:ieu-a-1004 0.1035706 0.002788592
correct_causal_direction steiger_pval
1 TRUE 0
5.9.0.1 Results for Reverse MR (Lipids → ANM)
5.9.0.2 LDL cholesterol (LDL → ANM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
-0.0865
0.1462
0.5563
Not significant
Weighted Median
-0.0693
0.0952
0.664
Not significant
IVW
-0.0945
0.0798
0.2361
Not significant
Simple Mode
-0.1890
0.2077
0.3662
Not significant
Weighted Mode
-0.1021
0.0844
0.2307
Not significant
As seen on the table above and MR scatter plot, the estimates are negative meaning higher LDL levels leads to earlier age at menopause. However all the values are insignificant.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 3.001923e-06 nd 4.335037e-06 respectively), indicating inconsistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.9476), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, ANM to LDL.
Single SNP analysis (and the forest plot) -> Most SNPs show insignificant (p-values >0.05 and CI crossing 0). The effect sizes go in both positive and negative directions. MR-Egger and IVW summary estimates overlap but also cross 0, thus not significant, indicating no specific outliers.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
5.9.0.3 HDL cholesterol (HDL → ANM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
-0.1490
0.1118
0.1865
Not significant
Weighted Median
-0.1198
0.0764
0.1169
Not significant
IVW
-0.0610
0.0707
0.3884
Not significant
Simple Mode
-0.2607
0.1632
0.1139
Not significant
Weighted Mode
-0.1209
0.0607
0.0497
Not significant (borderline)
As seen on the table above and MR scatter plot, the estimates are negative meaning higher LDL levels leads to earlier age at menopause. However all the values are insignificant.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 3.952637e-09 and 3.176397e-09 respectively), indicating inconsistent SNP effects and suggesting possible pleiotropic effects.
MR-Egger pleiotropy test -> The MR egger intercept was not significant (p-value 0.3130), indicating no evidence of directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, HDL to AAM.
Single SNP analysis (and the forest plots) -> Most SNPs show insignificant (p-values >0.05 and CI crossing 0). MR-Egger and IVW summary estimates overlap and close to 0, thus not significant. Although a few individual SNPs appear significant, these do not influence the overall causal estimate, and no single SNP appears to be driving the result.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis -> All p-values remained insignificant, suggesting no SNP is driving the observed effect.
LOO plot -> Each dot represents the IVW estimate after removing one SNP at a time. All the SNPs CI cross 0 and show have insignificant results, indicating no SNP shifts/drives the effect.
Conclusion: TC does not causally influence age at menopause.
5.9.0.4 Triglycerides (TG → ANM)
Method
β (Effect)
SE
p-value
Interpretation
MR-Egger
0.2535
0.1560
0.1102
Not significant
Weighted Median
0.1875
0.1093
0.0863
Not significant
IVW
0.2197
0.0940
0.0195
Significant
Simple Mode
0.2758
0.2324
0.2406
Not significant
Weighted Mode
0.0946
0.1204
0.4354
Not significant
As seen on the table above and MR scatter plot, the estimates are positive indicating higher TGs delays. The p-values is significant for IVW which is our main method, suggesting causal effect of TG levels on ANM.
Sensitivity analyses
Heterogeneity -> Significant heterogeneity was observed ( MR Egger and IVW Qp-values 4.609612e-07 and 6.721188e-07 respectively), indicating inconsistent SNP effects.
MR-Egger pleiotropy test -> The MR egger intercept was insignificant (p-value 0.7859), indicating evidence for no directional pleiotropy.
Steiger directionality test -> p = 0, significant, causal direction is correct, TG to ANM.
Single SNP analysis (and the forest plots) -> Most of the SNPs show insignificant (p-values >0.05 and CI crossing 0), indicating no specific outliers. The effect sizes go in both positive and negative direction. MR-Egger and IVW summary estimates overlap. IVW does not cross 0 showing significant estimates.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis(and LOO plot) -> Some of the leave-one-out estimates remained statistically significant and on the same side of the null, with confidence intervals not crossing 0. This indicates that the causal estimate is not driven by any individual genetic variant.
Conclusion: TG causally affect age at natural menopause.
5.9.0.5 Total cholesterol (TC → ANM)
Method
β
SE
p-value
Interpretation
MR-Egger
-0.0951
0.1814
0.6017
Not significant
Weighted Median
-0.1351
0.0923
0.1434
Not significant
IVW
-0.0821
0.0884
0.3535
Not significant
Simple Mode
-0.2502
0.1785
0.1648
Not significant
Weighted Mode
-0.1817
0.0977
0.0667
Not significant
As seen on the table above and MR scatter plot, the estimates lie close to 0 with some most showing positive effect sizes but the p-values are insignificant, suggesting no causal effect of TG levels to ANM.
Sensitivity analyses
Heterogeneity -> Borderline significant heterogeneity was observed ( MR Egger and IVW Qp-values 0.0005 and 0.0004 respectively), indicating some SNOs inconsistency.
MR-Egger pleiotropy test -> The MR egger intercept was insignificant (p-value 0.3322), indicating no directional pleiotropy.
Steiger directionality test -> p =4.391643e-271 significant, causal direction is correct, TC to ANM.
Single SNP analysis (and the forest plots) -> The SNPs show insignificant (p-values >0.05 and CI crossing 0). MR-Egger and IVW summary estimates overlap and close to 0, thus not significant.
Funnel plot: The SNPs show symmetrical distribution, indicating no directional horizontal pleiotropy.
Leave-one-out analysis(and LOO plot) -> Most of the leave-one-out estimates remained statistically insignificant and on the same side of the null, with confidence intervals crossing 0. This indicates that the causal estimate is not driven by any individual genetic variant.
Conclusion: TC does not influence age at natural menopause.
5.9.1Reverse Multivariable MR (MVMR)
Direction: Lipids + BMI → ANM
Exposures: HDL, LDL, TG and TC , Body Mass Index
Outcome: ANM
Research Question: Do lipid levels influence ANM independent of BMI? and does BMI affect the each lipid level independent of ANM.
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs7534572, rs9491696, rs9914578
Harmonising LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222) and Cholesterol, total || id:ebi-a-GCST002221 (ebi-a-GCST002221)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs7534572, rs9491696, rs9914578
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs7534572, rs9491696, rs9914578
Harmonising LDL cholesterol || id:ebi-a-GCST002222 (ebi-a-GCST002222) and Body mass index || id:ieu-a-2 (ieu-a-2)
Removing the following SNPs for being palindromic with intermediate allele frequencies:
rs14810, rs17001654, rs3758348, rs378479, rs9491696, rs9914578
id.exposure exposure id.outcome
1 ebi-a-GCST002216 Triglycerides || id:ebi-a-GCST002216 ieu-a-1004
2 ebi-a-GCST002221 Cholesterol, total || id:ebi-a-GCST002221 ieu-a-1004
3 ebi-a-GCST002222 LDL cholesterol || id:ebi-a-GCST002222 ieu-a-1004
4 ebi-a-GCST002223 HDL cholesterol || id:ebi-a-GCST002223 ieu-a-1004
5 ieu-a-2 Body mass index || id:ieu-a-2 ieu-a-1004
outcome nsnp b se pval
1 Age at menopause || id:ieu-a-1004 35 -0.1684315 0.3673050 0.6465500
2 Age at menopause || id:ieu-a-1004 67 0.9683181 0.9968569 0.3313634
3 Age at menopause || id:ieu-a-1004 53 -0.9215895 0.8679854 0.2883461
4 Age at menopause || id:ieu-a-1004 69 -0.4519502 0.4398711 0.3042037
5 Age at menopause || id:ieu-a-1004 51 -0.1394662 0.1631868 0.3927496
#Plot reverse MVMRggplot(res_mvmr_lipids_to_anm_bmi$result, aes(x = exposure, y = b)) +geom_point(size =3) +geom_errorbar(aes(ymin = b -1.96* se, ymax = b +1.96* se),width =0.15 ) +theme_bw() +theme(axis.text.x =element_text(angle =30, hjust =1, size =11) ) +labs(title ="Reverse MVMR: Effects of Lipids and BMI on Age at Natural Menopause (ANM)",x ="Exposure",y ="Effect on ANM" )
5.9.1.1 Results for Reverse MVMR (Lipids + BMI → ANM)
Exposure
SNPs
β
SE
p-value
Interpretation
Triglycerides (TG)
35
-0.1684
0.3673
0.6466
Not significant
Total Cholesterol (TC)
67
0.9683
0.9969
0.3314
Not significant
LDL Cholesterol
53
-0.9216
0.8680
0.2883
Not significant
HDL Cholesterol
69
-0.4520
0.4399
0.3042
Not significant
Body Mass Index (BMI)
51
-0.1395
0.1632
0.3927
Not significant
Conclusion: No evidence that increase in Lipid levels, even after adjusting for BMI influence the menopause timing
6Conclusion
This study investigated the causal links between reproductive timing—age at menarche (AAM) and age at natural menopause (ANM)—and major metabolic traits, including type 2 diabetes (T2D), lipid levels (HDL, LDL, triglycerides, total cholesterol), and BMI. Overall, the findings provide strong evidence that BMI is the key driver linking reproductive timing and metabolic risk, while AAM and ANM themselves show limited or trait-specific causal effects with the exception of AAM influencing HDL. Therefore, we can conclude that AAM and ANM are not reliable standalone predictors of T2D or lipid disorders when considered independently of BMI. BMI should be prioritized as a modifiable risk factor in the prevention of early puberty and metabolic disease. AAM may be useful as an early-life indicator of metabolic vulnerability, but mainly because it reflects underlying adiposity rather than reproductive biology. These findings suggest that reproductive timing reflects underlying metabolic risk largely through adiposity rather than direct hormonal mechanisms.
Limitations
Several multivariable Mendelian randomization analyses could not be performed due to insufficient SNP overlap following harmonisation. While forward Mendelian randomization analyses were successful, reverse and multivariable analyses required genetic variants to be simultaneously associated with multiple exposures and present in the outcome GWAS. Reproductive timing traits such as age at menarche and age at natural menopause are highly polygenic, whereas lipid traits and type 2 diabetes share fewer genome-wide significant variants with these traits. As a result, the intersection of valid instruments across all datasets was limited, leading to harmonisation failure. This reflects differences in genetic architecture rather than methodological error and highlights a known limitation of multivariable MR when applied to complex traits.
Strong heterogeneity in many models suggests biological complexity and possible balanced pleiotropy. Meaning that the genetic variants affect the outcome through other pathway than through the exposure, but these pleiotropic effects are random and cancel out in average. The results are specific to European-ancestry populations and it was difficult to find sex-stratified GWAS studies.
Future directions
Replicate in other ancestries to see whether the causal effects are the same as what we have seen in this study.
Explore the causal effects of reproductive timing with other cardiovascular traits and other metabolic traits.
Investigate the multivariable hormonal modeling.
Understand the biological mechanism of where we see positive causation.
Explore other pleiotropy and heterogeneity methods
MR-PRESSO - detects and eliminates outlier SNPs driving pleiotropy.